A new system architecture for applying symbolic learning techniques to robot manipulation tasks

摘要

Presents a new system architecture that enables a robot control system to learn at a symbolic level during planning and executing tasks. A user can program the robot by simply demonstrating the tasks it should perform. By performing inductive generalization and specialization steps, the system is able to improve its knowledge base. The architecture consists of a set of distributed knowledge units which realize a focus of attention that is necessary for efficient execution and learning, and which also makes competition between different problem solutions possible.